How Amazon’s AI Algorithms Raise the Prices You Pay 

In 2023, the Federal Trade Commission, then led by Joe Biden appointee Lina Khan, along with 17 states, filed an antitrust lawsuit against Amazon. The case, finally scheduled to go to trial next year, alleges that the Seattle-based behemoth maintains a monopoly in online retail through an interlocking set of tactics—what the complaint calls “a self-reinforcing cycle of dominance”—that thwart competitors and entrench its grip on the market. 

Central to Amazon’s monopoly power, the complaint alleges, are sophisticated AI-driven pricing systems that draw on torrents of real-time data and “can detect any price change virtually anywhere on the internet.” Amazon has used these algorithms not only to adjust its own prices in real time, but to monitor how competing retailers’ algorithms react, probe their strategies, and learn how to shape their behavior—including how to induce them to raise their prices.  

The FTC singles out what it calls Amazon’s ‘anti-discounting’ algorithm. According to the commission, it was conceived by Jeff Wilke, Amazon’s former head of Worldwide Consumer, to avoid what he described internally as a “perfectly competitive market,” in which rivals compete on price and drive down profits. Instead, Wilke argued for Amazon to adopt a “game theory approach,” predicting that if it did, both the company’s and its competitors’ prices “will go up.”  

Here’s how it allegedly worked: Amazon’s anti-discounting algorithm immediately matched competitors’ price changes to the penny, but never undercut them. When a rival offered a discount, Amazon’s algorithm matched it; when rivals raised prices, Amazon’s algorithm followed. This denied competing retailers a crucial tactic for luring customers from Amazon. If other retailers could never offer lower prices, Amazon’s roughly 200 million paying subscribers had little reason to shop elsewhere.   

More insidiously, the lawsuit contends, this strategy rewired how rivals price their products. The pricing algorithms widely used by online retailers are designed to learn and adapt; what they learned from Amazon was that price cuts yielded only thinner margins, not higher sales. Competitors’ algorithms eventually gave up on discounting and instead concluded that raising prices was a more profitable response. If sales volume could not grow, at least margins could. In turn, Amazon matched competitors’ price hikes, and prices across markets rose without any direct coordination. 

This strategy worked against the most well-resourced challengers. According to the complaint, in 2017, when Walmart sought to attract online shoppers by offering “pickup discounts” on about 1 million products available at retail outlets, Amazon instantly matched the price cuts. Walmart soon scaled back the program, “discount[ing] hundreds of thousands fewer products than it had initially planned.”  

Amazon also deployed its anti-discounting algorithm against Jet.com, according to the FTC. The company’s executives believed that Jet, a well-funded e-commerce startup that launched in 2015, could deliver “prices up to 10-15 percent lower.” But less than three months after Jet’s debut, Amazon’s tactics forced the newbie to abandon its price-leadership strategy and merely match Amazon’s prices. Walmart acquired Jet less than a year later and shut it down in 2020. The FTC’s complaint describes a similar trajectory for another e-commerce startup, Zulily, which, in 2019, began to display Amazon’s prices alongside its own lower ones. After Amazon targeted the upstart with its anti-discounting strategies, Zulily’s customer traffic plunged. (The company ceased operating and liquidated in 2023, though the daily-deals retailer Proozy has since purchased the brand and sought to revive it.) 

From time immemorial, businesses have paid attention to their competitors’ prices. But before today’s technologies, that information was partial and delayed. A retailer couldn’t see every rival’s price, everywhere, all the time.

Another algorithm—code-named Project Nessie—inverted this dynamic. Whereas the anti-discounting algorithm is reactive, Nessie is proactive: It learned to predict when Amazon’s rivals were likely to match its price hikes. Then it raised prices in those instances—allowing the company to charge more, knowing that shoppers were unlikely to find a cheaper option elsewhere. Like its mythical namesake, Nessie surfaced only when it could avoid detection: According to the FTC, Amazon paused Nessie during the holiday shopping season and Prime Day, when it feared customers or journalists might notice the price increases, then turned it back on once attention died down. The FTC alleges that Nessie generated over $1 billion in profit between 2016 and 2018. 

These allegations point to a novel form of monopoly power: the ability of a dominant platform to use algorithms to lift prices across an entire market. Amazon is doing so, the states and the FTC contend, not through explicit collusion, but by provoking and shaping the responses of rival algorithms, creating feedback loops that ratchet prices higher. If the FTC prevails, it will have proven that Amazon forged a monopolization strategy without precedent—one that exploited how algorithms have fundamentally altered pricing dynamics before the public and lawmakers knew to look. Amazon has denied the allegations, maintaining its pricing systems are designed to keep prices low and benefit consumers. 

Prices on the internet change constantly. To many shoppers, this volatility is a sign that digital markets are fiercely competitive. After all, comparison shopping is a click away, and retailers can monitor and undercut one another’s prices with ease. Such transparency, many assume, should drive prices lower.    

But the rapid price swings we see online often signal something quite different. Rather than reflecting competition or underlying market conditions, such as shifts in supply or demand, they’re more often artifacts of automated pricing systems reacting to one another. Today, most major online retailers use these tools, as do many third-party sellers operating on marketplaces run by Amazon and Walmart. These systems vary widely in their sophistication, speed, and breadth of real-time data they can access. Some follow simple rules; others use machine learning to learn, strategize, and adapt.  

From time immemorial, businesses have paid attention to their competitors’ prices. But before today’s technologies, that information was partial and delayed. A retailer couldn’t see every rival’s price, everywhere, all the time. Nor could it react instantly. That friction created opportunities for challengers. A grocery store looking to draw customers from a dominant supermarket might slash prices on popular products like milk, ground beef, or Cheerios. The discounts would bring shoppers through the door, where they might notice other aspects of the store, such as its great service or well-stocked produce section. Even if the larger supermarket matched the smaller rival’s lower prices within a day or two, the smaller rival could pick up new long-term customers. 

Shoppers correctly perceive the dominant retailer as the low-price leader and flock to it. What they cannot see is how its algorithmic strategies are quietly ratcheting prices upward—both on the platform’s own products and across the market. 

Pricing algorithms eliminate the time and information frictions that once shaped pricing decisions. In a genuinely competitive online market, one in which customer traffic was widely dispersed across many retailers, such transparency might intensify price competition. But when a single company commands an outsized share of the market—Amazon captures as much as 50 percent of online sales in the U.S. and an even larger share in some categories—these technologies can become tools for price manipulation. 

Amazon’s pricing algorithms are regarded by industry observers as the most advanced on the web (“best-in-class” according to Bain, the consulting firm). The company’s market dominance gives its systems an added edge. Because rivals are trying to peel away Amazon’s customers, their algorithms, out of necessity, are heavily oriented toward Amazon’s pricing. This allows Amazon to use competitor responses as information signals: It can test how quickly they react to its price changes, how closely they match a newly set price, and whether they follow increases as readily as they follow cuts.  

Over time, a dominant firm can learn how to shape the pricing behavior of its competitors. Economists Zach Brown and Alexander MacKay call this phenomenon “algorithmic coercion.” In their models, a dominant company equipped with faster, more advanced pricing technology can induce rival firms to raise prices. The algorithms most effective at doing so are those with the broadest market visibility, the fastest reaction times, and the strongest strategic commitment—meaning a willingness to repeat a pricing response consistently, even if it entails short-term losses. Such behavior disciplines rival algorithms, which stop cutting prices and begin raising them instead, just as allegedly happened with Amazon’s competitors. As Brown and MacKay write, a superior algorithm unilaterally pursuing this strategy “can be worse for consumers than if firms explicitly coordinated on high prices.” 

Brown and MacKay tested this theory using more than a year’s worth of hourly pricing data for seven over-the-counter allergy drugs sold by the five largest online retailers in the category. The companies’ algorithms varied widely in sophistication, with one clearly more advanced than the rest. (The researchers do not name the firms.) Applying their model to the data, they found that algorithmic pricing increased the average prices for the seven drugs by 5 percent. The biggest winner, of course, was the retailer with the most capable algorithm. It not only raised its own prices but also gained market share by inducing rivals to raise theirs even more.  

The result is a troubling inversion of the way price competition is meant to work. Shoppers correctly perceive the dominant retailer as the low-price leader and flock to it. What they cannot see is how its algorithmic strategies are quietly ratcheting prices upward—both on the platform’s own products and across the market.  

Amazon shapes pricing decisions not only across the web, but within its own vast third-party marketplace. Third-party sellers, which account for about 60 percent of the platform’s sales, are a diverse lot, but they all depend on Amazon’s infrastructure. Most have no viable alternative to selling on Amazon, and the platform’s algorithmic systems largely control their fate—determining where they appear in search results and dispensing punishments, including suspensions, for perceived infractions. 

To survive in Amazon’s ecosystem, most sellers also rely on algorithms. Some use Amazon’s basic software, but many turn to third-party “repricers.” These tools function less as aids to competing and more as adaptations to a marketplace structured by a dominant gatekeeper. Sellers deploy them in hopes of navigating Amazon’s opaque ranking algorithms while keeping prices high enough to offset the platform’s rising fees. Unsurprisingly, repricing firms market their ability to fatten profits. One of the most popular, Aura, claims it “combines AI-driven pricing with 10-second repricing speeds” to enable sellers to “avoid price wars” and “maximize profitability.” Another, BQool advertises that its algorithm can “automatically detect … a chance to raise the price.” 

Repricers exploit Amazon’s platform rules. For popular products offered by multiple sellers at similar prices, Amazon rotates the Buy Box—the default seller position—giving each seller a turn. According to the economist Ted Tatos, repricers have pioneered a “wait my turn” strategy: Instead of undercutting the seller holding the Buy Box, they hold back, allowing that seller to raise prices, knowing they will soon get to sell at the same higher price. SellerSnap, another popular repricer, touts this “cooperative” approach. “A seller’s instinct might be to lower their price to win more of the Buy Box,” the company notes, but “by making use of a cooperative strategy, a seller can secure their fair Buy Box share, avoiding price wars and maximizing profit.” Its tool automates this behavior.  

Daily price cycling is another tactic repricers use to lift prices. During low-traffic overnight hours, repricers sharply reset prices upward to see if rivals follow. If so, the price floor for that product rises. In an analysis of pricing data for thousands of products sold on Amazon between 2019 and 2020, economist Leon Musolff found a recurring pattern: Prices spike overnight, then slowly erode as sellers resume undercutting one another. Over time, the cycle “coaxes competitors to raise their prices,” pushing average prices up by more than 11 percent. This reset strategy was relatively rare in 2020, Musolff found. Today, repricing firms openly market it. SellerSnap’s software offers a “Yo-Yo Repricing Rule,” while Aura promotes an “Oscillation Strategy.” 

Amazon has no incentive to stamp out these practices. It takes 45 percent of every third-party sale in the U.S. through referral, fulfillment, and advertising fees. These seller fees are highly lucrative; the company generated $125 billion from U.S. third-party sellers in 2023. Amazon safeguards this revenue stream through a third pricing algorithm at issue in the FTC’s lawsuit (and which is also targeted by separate antitrust cases brought by California, D.C., and Arizona). Dubbed the Featured Offer Disqualification algorithm, it monitors prices across the web and punishes any seller that offers a lower price elsewhere by making them ineligible to win the Buy Box, effectively wiping out their sales. This strategy ensures that, as Amazon’s fees rise, sellers recoup those costs by increasing their prices not only on Amazon but across the web. 

State and federal lawmakers increasingly worry that algorithmic pricing technologies might be used to exploit consumers. But so far, policy efforts have mainly targeted two well-known types of abuse.  

One involves surveillance or personalized pricing: The use of detailed customer data to charge different prices for the same product. A recent investigation, for example, found that Instacart charged grocery shoppers widely varying prices for identical items purchased at the same time from the same store. Several states are considering legislation to limit such practices. New York already requires companies that adjust prices based on individual data to disclose that fact to customers; state lawmakers are now considering a measure that would ban the practice. Meanwhile, Maryland lawmakers are debating a proposal that would prohibit surveillance pricing by grocery retailers, including both online sellers and physical stores, some of which are beginning to deploy digital shelf tags, allowing them to change prices with the ease and frequency of online retailers. 

A second area of concern: Rivals using software to collude. RealPage, for example, has been accused by antitrust enforcers of facilitating a cartel among major landlords, allowing them to raise rents in lockstep. In response, several cities and states have passed laws banning pricing tools that pool data from competing property owners and make recommendations for how much to charge in rent. A new law in California extends this prohibition to other sectors of the economy beyond housing. 

Yet these emerging policy approaches do not grapple with the kind of unilateral algorithmic price manipulation Amazon is accused of. Both new regulations and existing law governing collusion generally hinge on evidence that companies relied on the same software, shared non-public information, or otherwise coordinated indirectly. Amazon’s pricing algorithms, by contrast, rely on none of these mechanisms. They draw from publicly posted prices and learn through a continuous call-and-response with rival algorithms—without any shared software, confidential data, or coordination. Instead, they use their own price changes to provoke responses that create feedback loops, pushing prices higher across the market. 

Policymakers need to regulate not just how prices are set, but how often they can change. One approach would be to limit the frequency of price changes to once a day. Such a rule would prevent the rapid reactions that allow algorithms to orchestrate parallel price moves.

Even California’s new law—a breakthrough, in that the sharing of non-public information is not a required element—still targets only situations in which a platform coerces sellers to use the same pricing tool. It’s therefore unlikely to affect Amazon, since both the company’s external competitors and its internal third-party sellers rely on a wide variety of automated pricing systems. Moreover, as Brown and MacKay’s research finds, the asymmetry of these systems, in terms of technological sophistication and speed, actually enhances a dominant firm’s ability to shape market-wide pricing decisions.  

What the case against Amazon raises—and what lawmakers have largely failed to grapple with—is the possibility that a dominant platform equipped with powerful pricing technology may be able to raise prices across a market without colluding in any conventional way.  

To address this, policymakers need to regulate not just how prices are set, but how often they can change. One approach would be to limit the frequency of price changes to once a day. Such a rule would prevent the rapid reactions that allow algorithms to orchestrate parallel price moves. Introducing a delay would also restore the frictions that once enabled challengers to gain a foothold through discounting. And it would push companies to offer their best prices each day, rather than risk posting prices significantly higher than their rivals. An early version of Maryland’s grocery bill included such a provision, but it was later stripped out. 

In the absence of measures along these lines, the problem is likely to intensify. As Amazon develops more powerful AI tools, its capacity to shape rivals’ prices is poised to accelerate. With more advanced pattern recognition and learning capabilities—and the ability to gather and process ever larger streams of real-time data—its systems will likely develop increasingly sophisticated strategies for influencing market-wide pricing. In 2026 alone, Amazon plans to invest a staggering $200 billion in AI infrastructure and development. 

Allowing algorithmic pricing to spread unregulated poses an even greater danger than higher prices: It threatens to undermine the core functionality of markets. Markets work by conveying information—quickly and broadly—through prices. Rising prices signal strong demand and invite new competition and production. Falling prices indicate slack or waste and prompt resources to be reallocated. Innovation often flows from price signals too, as higher input costs spur businesses to develop alternatives. 

But when prices are set by algorithms driven less by real-world conditions like supply and demand and more by the pricing moves of rival algorithms, those signals degrade. Price shifts cease to reflect underlying economic conditions and instead become noisy artifacts of algorithms reacting to one another. As such, prices no longer function as reliable carriers of information and guides for decisions about investment and production. While policymakers have begun to recognize that algorithms can harm consumers, they’ve yet to contend with this more fundamental risk.  

This is why so much may turn on the government’s case against Amazon, which is scheduled to go to trial next March. Seventeen states have joined the case, and could carry it forward even if the Trump administration backs away (as it has in other antitrust matters). The lawsuit survived a motion to dismiss, with the judge holding that the FTC and states plausibly alleged that Amazon’s algorithms “stifle price competition.” Notably, his ruling marks the first time a court has held that systematically tracking and responding to rivals’ prices, absent an allegation of collusion, can violate the Sherman Antitrust Act. The judge also allowed the government’s claim that Project Nessie constitutes an “unfair method of competition” under the FTC Act to proceed as a standalone violation, without requiring proof of antitrust injury under the Sherman Act’s more restrictive standards.   

If the government succeeds on these points, antitrust law may finally begin to catch up to how digital markets work. If not, the volatility that greets shoppers online will continue to masquerade as competition, even as it signals something else: A market in which prices still move, but no longer perform the work we depend on them to do.

The post How Amazon’s AI Algorithms Raise the Prices You Pay  appeared first on Washington Monthly.